tag:blogger.com,1999:blog-6141980.post5789905265258546300..comments2024-03-20T12:28:35.004-05:00Comments on Nuit Blanche: Sunday Morning Insight: Faster Than a Blink of an EyeIgorhttp://www.blogger.com/profile/17474880327699002140noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-6141980.post-1379779710212506092013-07-10T05:17:36.210-05:002013-07-10T05:17:36.210-05:00Thanks Leslie.
Igor.Thanks Leslie.<br /><br />Igor.Igorhttps://www.blogger.com/profile/17474880327699002140noreply@blogger.comtag:blogger.com,1999:blog-6141980.post-54793346993606007092013-07-09T06:03:12.546-05:002013-07-09T06:03:12.546-05:00Excellent post; you crystallized some of my own th...Excellent post; you crystallized some of my own thoughts on this topic. Let me add a few other relevant thoughts. <br /><br />You wrote "I wonder how, instead of ISTA/FISTA, an AMP solution would provide an even faster capability since we know these iterations schemes to be pretty fast." <br /><br />My experience with sparse coders like ISTA/FISTA is that a warm start (i.e., a good guess of the solution) will reduce the number of needed iterations, hence speed up the sparse coding. When I read your [5], my thought was that the neural net did just that - provide a good guess and reducing the number of iterations by an order of magnitude.<br /><br />Although I don't have direct experience with AMP, my understanding of the AMP literature is that few iterations are required. Hence I was thinking that in this case the neural net would not provide much speed improvement. But I could be wrong about this and would like to hear from an "AMP practitioner" as to the impact of a neural net or other good solution estimator on AMP.<br /><br />Thanks again for this blog posting on this interesting topic.Anonymoushttps://www.blogger.com/profile/06179125920265282125noreply@blogger.com